Near infrared hyperspectral imaging and chemometrics for exploration and classification of whole wheat kernels

dc.contributor.advisorManley, Marena
dc.contributor.advisorGeladi, Paul
dc.contributor.authorDu Toit, Gerida
dc.contributor.otherUniversity of Stellenbosch. Faculty of Agrisciences. Dept. of Food Science.
dc.date.accessioned2009-11-26T06:54:26Zen_ZA
dc.date.accessioned2010-06-01T09:02:58Z
dc.date.available2009-11-26T06:54:26Zen_ZA
dc.date.available2010-06-01T09:02:58Z
dc.date.issued2009-12
dc.descriptionThesis (Msc Food Sc (Food Science))--University of Stellenbosch, 2009.
dc.description.abstractENGLISH ABSTRACT: Near infrared (NIR) hyperspectral imaging together with multivariate image analysis was evaluated as a non-destructive method to distinguish between whole wheat kernels differing in hardness; and also to track the diffusion of conditioning water into whole wheat kernels of different hardness over a conditioning period of 36 hours. Wheat kernels of varying hardness were imaged using a Spectral Dimensions MatrixNIR imaging system with a wavelength range of 960-1662 nm. Principal component analysis (PCA) was applied to clean the image, which entailed removal of bad pixels (background, shading, curvature errors, dead pixels and outliers). PCA also proved effective in the identification and classification of clusters in the score plot, relating to different hardness wheat endosperm (durum, hard and soft). PC 2 differentiated soft endosperm form hard and durum endosperm; while PC 3 distinguished durum endosperm from hard and soft endosperm. The loading line plot of PC 2 indicated absorbance peaks at 1195, 1450 and 1570 nm associated with starch, moisture and protein; while the loading line plot of PC 3 indicated absorbance peaks at 1195 and 1450 nm associated with starch and moisture. Partial least squares discriminant analysis (PLS-DA) was used to determine the ability to discriminate between different hardness endosperm classes using NIR hyperspectral imaging. The model of soft versus (vs) durum endosperm obtained a classification accuracy of 100%; the model of soft vs hard endosperm 98% classification accuracy; and the model of hard vs durum endosperm model classification accuracy up to 96%. NIR hyperspectral images were acquired using the sisuChema SWIR (short wave infrared) imaging system with a wavelength range of 1000 to 2500 nm. Images of wheat conditioned with water (H2O) and deuterium oxide (D2O), respectively, were acquired at regular intervals between 0 and 36 hours. PCA proved effective in cleaning the image. The score images of PC 3 for wheat conditioned with H2O indicated an increase in intensity over conditioning time. The loading line plots of PC 3 for wheat conditioned with H2O indicated the variation in PC 3 due to bound moisture (1940 nm). Comparing the results from the score images and loading line plots, a conclusion could be made that the diffusion of conditioning water into soft wheat kernels reaches equilibrium after 18 hours, 24 hours for hard wheat and 36 hours for very hard wheat. The score images of wheat conditioned with D2O indicated an increase in intensity within either PC 3 or PC 5; intensity increases were between 0 and 6 hours with no further increase up to 36 hours conditioning. The loading line plots of PC 3 and PC 5 indicated variation in these PCs due to D2O (1954 nm). In contrast to results obtained with H2O, D2O did not diffuse into the wheat endosperm as expected. NIR hyperspectral imaging proved effective in differentiating between whole wheat kernels differing in hardness; and also in tracking the diffusion of conditioning water into whole wheat kernelsen
dc.description.abstractAFRIKAANSE OPSOMMING: Die gebruik van naby infrarooi (NIR) hiperspektrale beelding en veelvoudige beeldanalise is beoordeel as ‘n nie-destruktiewe metode om onderskeid te tref tussen heel koringkorrels van verskillende hardhede, sowel as die volg van die diffusie van aanklammingswater in heel koringkorrels van verskillende hardhede oor ‘n periode van 36 uur. NIR hiperspektrale beelde is verkry van verskillende hardhede koringkorrels deur gebruik te maak van ‘n Spectral Dimensions MatrixNIR kamera met ‘n spektrale reikwydte van 960-1662 nm. Hoofkomponent-analise (HKA) is toegepas om die beeld skoon te maak. Skoonmaak van die beeld het ingesluit die verwydering van agtergrond, skadu, krommingsfoute en dooie piksels. HKA is doeltreffend gebruik vir die identifikasie en klassifikasie van histologiese klasse naamlik durum-endosperm, harde-endosperm en sagte-endosperm. Hoofkomponent (HK) 2 het duidelike onderskeid getref tussen sagte-endosperm en hard- en durum-endosperm; terwyl HK 3 onderskeid tussen durum-endosperm en harde- en sagte-endosperm aangedui het. Die HK lading-stip van HK 2 het absorpsiepieke by 1195, 1450 en 1570 nm aangedui wat met stysel, vog en proteïen geassosieer kon word. Die lading-stip van HK 3 het absorpsiepieke by 1195 en 1450 nm aangedui wat verband hou met stysel en vog. Parsiële kleinste waarde diskriminant-analise (PKW-DA) is gebruik om die moontlikheid van diskriminasie tussen verskillende hardhede koring vas te stel deur van NIR hiperspektrale beelding gebruik te maak. Die model van sagte- teenoor durum-endosperm het ‘n klassifikasie koers van 100% bereik, die model van sagte- teenoor harde-endosperm ‘n klassifikasie koers van 98%; en die model van harde- teenoor durum-endosperm ‘n klassifikasie koers van 96%. Beelde van koring, aangeklam met water (H2O) en deuteriumoksied (D2O), onderskeidelik, is verkry deur gebruik te maak van die sisuChema SWIR (short wave infrared) kamera met ‘n spektrale reikwydte van 1000-2500 nm. Beelde van die aangeklamde koring is by gereelde intervalle oor ‘n periode van 36 uur verkry. HKA kon effektief gebruik word om die beeld skoon te maak. Die telling-beeld van HK 3, vir koring met H2O aangeklam, het ‘n toename in intensiteit oor die aanklammingstydperk getoon, terwyl die lading-stip van HK 3 aangedui het dat die variasie in die HK aan gebonde vog (1940 nm) toegeskryf kon word. Deur die resultate van die lading-stip en telling-stip te vergelyk kon daar tot ‘n gevolgtrekking kom dat die diffusie van aanklammingswater in sagte korrels na 18 uur ‘n ekwilibrium bereik het, terwyl die ekwilibrium na 24 uur in harde en 36 uur in baie harde korrels bereik is. Die telling-beelde van koring, aangeklam met D2O, het ‘n toename in intensiteit aangedui in HK 3 of HK 5, onderskeidelik. Die intensiteit in die telling-beeld het toegeneem vanaf 0 tot 6 uur na aanklamming, waarna daar geen verdere toename in intensiteit tot en met 36 uur was nie. Die HK lading-stip van HK 3 en HK 5 het aangedui dat die variasie in hierdie hoofkomponente aan D2O (1954 nm) toegeskryf kon word. In teenstelling met die resultate verkry van die aanklamming met H2O, het D2O geen diffusie binne die korrel aangedui nie. Die gebruik van NIR hiperspektrale beelding was suksesvol om onderskeid te tref tussen heel koringkorrels van verskillende hardhede, en ook met die volg van die diffusie van aanklammingswater in heel koringkorrelsaf
dc.identifier.urihttp://hdl.handle.net/10019.1/2978
dc.language.isoen
dc.publisherStellenbosch : University of Stellenbosch
dc.rights.holderUniversity of Stellenbosch
dc.subjectWheat kernel hardnessen
dc.subjectWheat conditioningen
dc.subjectHyperspectral imagingen
dc.subjectDissertations -- Food scienceen
dc.subjectTheses -- Food scienceen
dc.subject.lcshWheat -- Seedsen
dc.subject.lcshNear infrared spectroscopyen
dc.subject.lcshChemometricsen
dc.titleNear infrared hyperspectral imaging and chemometrics for exploration and classification of whole wheat kernelsen
dc.typeThesis
Files
Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
dutoit_near_2009.pdf
Size:
3.64 MB
Format:
Adobe Portable Document Format
Description: